Augmented Cycle GAN

A Theano re-implementation of Augmented CycleGAN

Requirements

Theano

Neuralnet

Usage

python train.py

Use --help for more options.

To test a pretrained network

python train.py --test --param_file_version a-saved-weight-file-version

Results

Sketches to shoes

Shoes to sketches

Sketches to multiple shoes

References

If you use the implementation, you must cite the following paper

@article{almahairi2018augmented,
  title={Augmented CycleGAN: Learning Many-to-Many Mappings from Unpaired Data},
  author={Almahairi, Amjad and Rajeswar, Sai and Sordoni, Alessandro and Bachman, Philip and Courville, Aaron},
  journal={arXiv preprint arXiv:1802.10151},
  year={2018}
}

Official implementation.